Goal-Conditioned Reinforcement Learning with Extended Floyd-Warshall method